Mask Aware Network for Masked Face Recognition in the Wild

Abstract

Face recognition is one of the most important research topics for intelligence security system, especially in the COVID-19 era. Medical research has proven that wearing a mask is the most efficient way to avoid the risk of COVID-19. Nevertheless, classic face recognition systems often fail when dealing with masked faces. So it is essential to design a method that is robust to Masked Face Recognition (MFR). In this paper, to relieve the degraded performance of MFR, we propose Mask Aware Network (MAN) including a mask generation module and a loss function searching module. The mask generation module utilizes the face landmarks to obtain more realistic and reliable masked faces for training. The loss function searching module tries to match the most suitable loss for face recognition. On ICCV MFR challenge, our team victor-2021 achieves 5 first places (including 3 champions in standard face recognition and 2 champions in masked face recognition) and 1 third place by 3rd August 2021. These results demonstrate the robustness and generalization of our method in both standard or masked face recognition task.